On a global scale, fire disasters present a serious risk to human well-being and the natural surroundings and infrastructure. Traditional fire detection methods encounter a challenge of data imbalance, with an abundance of non-fire scenario data but limited availability from actual fire incidents, resulting in challenges in building an accurate model capable of detecting various types of fires. Moreover, in real-life situations, sensor data collected from the non-fire scenario may exhibit patterns that fall into more than one class. This specific scenario type, known as nuisance scenarios (e.g., smoking and cooking), shares some characteristics with fire scenarios but is considered non-fire, and neglecting such cases can result in delayed fire detection or trigger nuisance alarms. To address this issue, this paper proposes a novel fire monitoring system based on a multi-class support vector data description with a dynamic time warping kernel, which considers distinct sensor patterns from multiple non-fire scenarios. The developed system is able to effectively detect different fire types without prior knowledge by constructing distinct decision boundaries for different non-fire classes. In addition, the proposed system further considers the temporal dynamic and shape patterns present within sensor signals by incorporating the dynamic time warping kernel function. Experimental results demonstrate the proposed system’s superiority over existing methods in detecting different fire types earlier and more accurately with lower false alarm rates in non-fire scenarios.

This paper addresses semi-supervised anomaly detection in settings where only a small subset of normal data can be labeled. Such conditions arise, for example, in industrial quality control of windshield wiper noise, where expert labeling is costly and limited. Our objective is to learn a one-class decision boundary that leverages the geometry of unlabeled data while remaining robust to contamination and scarcity of labeled normals. We propose a graph-attention-regularized deep support vector data description (GAR-DSVDD) model that combines a deep one-class enclosure with a latent k-nearest-neighbor graph whose edges are weighted by similarity- and score-aware attention. The resulting loss integrates (i) a distance-based enclosure on labeled normals, (ii) a graph smoothness term on squared distances over the attention-weighted graph, and (iii) a center-pull regularizer on unlabeled samples to avoid over-smoothing and boundary drift. Experiments on a controlled simulated dataset and an industrial windshield wiper acoustics dataset show that GAR-DSVDD consistently improves the F1 score under scarce label conditions. On average, F1 increases from 0.78 to 0.84 on the simulated benchmark and from 0.63 to 0.86 on the industrial case study relative to the best competing baseline.

Anomalies in real systems differ widely in impact, as such, missing a high-severity event can be far costlier and consequential than flagging a benign outlier. This paper introduces Severity-Regularized Deep Support Vector Data Description, an extention of deep support vector data description that incorporates severity for various anomaly types, reflecting the application-specific importance assigned to each type. The formulation retains the well-known deep support vector data description decision geometry and scoring system while allowing for specific control over the balance between false alarm rate and the prioritization of detecting anomalies with greater impact. In the proposed loss function, we introduce regularizing parameters that control the importance assign to each anomaly type. Experiments are carried out on a demanding simulated dataset and a real-world intrusion detection case study utilizing the Australian Defence Force Academy Linux Dataset. The results demonstrate the effectiveness of the proposed approach in detecting highly severe anomalies while maintaining competitive overall performance.

Anomaly detection in safety-critical systems often operates under severe label constraints where only a small subset of normal and anomalous samples can be reliably annotated, while large unlabeled data streams are contaminated and high-dimensional. Deep one-class methods, such as deep support vector data description (DeepSVDD) and deep semi-supervised anomaly detection (DeepSAD), address this setting. However, they treat samples largely in isolation and do not explicitly leverage the manifold structure of unlabeled data, which can limit robustness and interpretability. This paper proposes Anomaly-Aware Graph-based Semi-Supervised Deep Support Vector Data Description (AAG-DSVDD), a boundary-focused deep one-class approach that couples a DeepSAD-style hypersphere with a label-aware latent k-nearest neighbor (k-NN) graph. The method combines a soft-boundary enclosure for labeled normals, a margin-based push-out for labeled anomalies, an unlabeled center-pull, and a k-NN graph regularizer on the squared distances to the center. The resulting graph term propagates information from scarce labels along the latent manifold, aligns anomaly scores of neighboring samples, and supports sample-level explanations through graph neighborhoods, while test-time scoring remains a single distance-to-center computation. On a controlled two-dimensional synthetic dataset, AAG-DSVDD achieves a mean F1-score of 0.88±0.02 across ten random splits, improving on the strongest baseline by about 0.12 absolute F1. In a multi-sensor fire monitoring case study, AAG-DSVDD reduces the average fire starting time accuracy to approximately 473 seconds (about 30% improvement over DeepSAD), while keeping the average pre-fire false-alarm rate below 1% and avoiding persistent pre-fire alarms. These results indicate that graph-regularized deep one-class boundaries offer an effective and interpretable framework for semi-supervised anomaly detection under realistic label budgets.